KamLAND — the night reactors taught neutrinos to oscillate

By the late 1990s the solar neutrino problem had narrowed to a handful of solutions on a two-dimensional parameter plane: a small-mixing-angle MSW solution, a low-mass MSW solution, a vacuum-oscillation solution at much smaller splittings, and the large-mixing-angle MSW solution — known as LMA — with a mass splitting around 7 × 10⁻⁵ eV² and a near-maximal mixing angle. SNO and Super-Kamiokande had established that solar electron neutrinos really do change flavour on their way out of the Sun, but the precision was not yet enough to pick one of these regions over the others with high confidence. To kill the alternatives, what was needed was a man-made source with a precisely known baseline and a well-understood spectrum, sitting at exactly the L/E that LMA predicted should produce a large effect.

That is what nuclear reactors deliver. Each fission of uranium-235 or plutonium-239 produces roughly six electron antineutrinos as the fission fragments beta-decay toward stability. A gigawatt-thermal reactor radiates about 2 × 10²⁰ ν̄_e per second isotropically, with a spectrum well below 10 MeV that has been computed and measured for decades. At ~180 kilometres baseline and a few MeV of energy, the LMA parameters predicted an oscillation phase of order unity and an antineutrino flux suppression of roughly 40%. Japan is the country, more than any other, with the right geography for the experiment: dozens of commercial reactors scattered along its coasts, providing a flux-weighted average baseline of about 180 km to a detector placed under the same mountain that already housed Super-Kamiokande.

What KamLAND actually was

The Kamioka Liquid-scintillator Anti-Neutrino Detector was a one-kilotonne spherical balloon of mineral oil and pseudocumene scintillator, suspended in a stainless-steel vessel and watched by 1879 photomultipliers, all sitting in the Mozumi mine in the Japanese Alps. The detection reaction was inverse beta decay, ν̄_e + p → n + e⁺, the same workhorse that Cowan and Reines used in 1956. The signature is a tight delayed coincidence: a prompt flash from the positron (carrying most of the neutrino's energy minus 1.8 MeV), followed about 200 microseconds later by a 2.2 MeV gamma from the neutron capturing on a free proton. The double signature suppresses backgrounds by factors of thousands, allowing KamLAND to see a signal of only about one antineutrino interaction per day in a multi-kilotonne fiducial volume.

The result came in two stages. In December 2002, with 162 ton-years of exposure, the collaboration counted 54 candidate antineutrino events where 86 ± 6 were expected from the no-oscillation prediction — a deficit at 99.95% confidence and the first observation of reactor antineutrinos disappearing over a long baseline. By 2005, with 766 ton-years, there was enough statistics to bin the events in L/E and plot the survival probability bin by bin. That plot — the famous "KamLAND L/E plot" — is reproduced in textbooks for a reason. It shows the data tracking, with no fitting, the sinusoidal oscillation predicted by quantum mechanics across more than two full cycles in L/E. It was the first time anyone had seen the oscillation pattern itself with a controlled source, not merely inferred an averaged-out deficit.

Slide Δm²₂₁ around and watch the oscillation pattern slide horizontally. At the measured value of roughly 7.5 × 10⁻⁵ eV² the first minimum lands at about L/E = 33 km/MeV — right in the middle of the L/E range KamLAND covered — and a second maximum rises near 65 km/MeV. Both features were visible in the 2005 binned data. Push the slider far below the measured value and the first minimum walks off to the right, beyond the data; push it far above and the oscillations crowd together and would average out before KamLAND's resolution could resolve them. The fact that the data hit one specific value of Δm²₂₁, and no other, is what made the measurement of the solar mass splitting a matter of percent-level precision rather than order-of-magnitude estimation.

Why this experiment closed a chapter

Before KamLAND, the LMA region was one of several solutions to the solar neutrino problem and the favoured one only by Occam's razor — solar data alone could not exclude the alternatives at the level needed to settle the question. After KamLAND, the LMA region was the only solution. The vacuum and low-mass options were excluded outright by the wrong oscillation pattern; the small-mixing-angle solution was excluded by the depth of the deficit. The combined fit of all solar experiments and KamLAND, completed shortly afterward, pinned Δm²₂₁ to within about 3% and θ₁₂ to within a couple of degrees. The two solar oscillation parameters had moved from a list of candidates to a measurement.

This mattered far beyond the solar problem. The LMA solution requires neutrino flavour to be transformed by the matter effect — the MSW resonance — inside the Sun, which means that the sign of Δm²₂₁ and the structure of the PMNS matrix are not mere phenomenological labels but physical features of how a neutrino propagates through matter. It also means that the relevant solar mixing angle is large but not maximal, which is exactly the condition required for the observable CP violation that the long-baseline programme is now chasing. KamLAND's measurement, together with the Daya Bay measurement of θ₁₃ a decade later, completed the basic geometry of three-flavour neutrino oscillation: every angle and both mass splittings known to within a few percent, leaving only the ordering, the CP phase and the absolute mass scale to be measured by the next generation. The night the reactors taught neutrinos to oscillate was, in retrospect, the night the field stopped arguing about the parameter map and started using it.

For context on the master variable that makes all of this possible, see L over E — the master variable of oscillation; for the heavy-water experiment that established flavour change in the solar sector in the first place, see SNO's heavy-water gambit.